COSMOS DSL is a language dedicated to the domain of composition of context information. It allows, through its high-level constructions, to treat very large amounts of context data coming from distributed sources, whatsoever in ubiquitous systems or large-scale distributed systems. In this paper, we focus on the safety of programs developed in COSMOS DSL. We discuss the needs of reliability and validation of specific properties such as the lack of deadlocks and vivacity of context information. Then we propose an automatable modeling of COSMOS DSL using the Petri nets formalism to allow verification of behavioral properties in a software tool